Table 1.

Predictive accuracy of hospital admission and use of corticosteroids of different definitions of exacerbation.

Description		Practical	AUCa (empirical 95% CI)	Events, N+	Samples, N
Prediction of 24-hour admissions using exacerbation definitions, complete data
	Definition 1	Yes	0.600 (0.509-0.692)	17	17,610
	Definition 2	Yes	0.578 (0.496-0.672)	17	17,610
	Definition 3	No	0.553 (0.440-0.666)	8	14,106
	Definition 4	No	0.490 (0.424-0.556)	8	14,106
	Definition 5	No	0.657 (0.523-0.792)	9	16,170
Prediction of 24-hour admissions using exacerbation definitions, imputed data
	Definition 1	Yes	0.513 (0.477-0.551)	55	57,150
	Definition 2	Yes	0.524 (0.486-0.544)	55	57,150
	Definition 3	No	0.496 (0.471-0.521)	55	56,702
	Definition 4	No	0.505 (0.473-0.536)	55	56,702
	Definition 5	No	0.517 (0.479-0.555)	55	57,150
Prediction of 24-hour corticosteroid decisions using exacerbation definitions, complete data
	Definition 1	Yes	0.655 (0.630-0.679)	238	9768
	Definition 2	Yes	0.605 (0.581-0.628)	238	9768
	Definition 3	No	0.568 (0.544-0.592)	178	8489
	Definition 4	No	0.544 (0.522-0.567)	178	8489
	Definition 5	No	0.646 (0.622-0.670)	237	9322
Prediction of 24-hour corticosteroid decisions using exacerbation definitions, imputed data
	Definition 1	Yes	0.660 (0.639-0.681)	316	13,899
	Definition 2	Yes	0.605 (0.585-0.625)	316	13,899
	Definition 3	No	0.564 (0.543-0.586)	228	10,442
	Definition 4	No	0.543 (0.524-0.564)	228	10,442
	Definition 5	No	0.647 (0.626-0.668)	316	12,477
Prediction of 24-hour admissions using machine learning models, imputed data
	Machine learning model	Yes	0.740 (0.673-0.803)	55	57,150
Prediction of 24-hour corticosteroid decisions using exacerbation definitions, imputed data
	Machine learning model	Yes	0.765 (0.738-0.791)	316	13,503

^aAUC: area under the receiver operating characteristic curve.